An Overview of Software Development Effort and Cost Estimation Techniques

1
An Overview of Software Development Effort and
Cost Estimation Techniques
Professor Ron Kenett Tel Aviv University School
of Engineering
2
Software Cost Estimation Models

• TRW
• Doty
• Boeing
• IBM-FSD
• Rayleigh
• SLIM
• RCA Price/S
• COCOMO 81
• JPL

3
Software Cost Estimation Models

• Doty - 14 factors
• IBM-FSD - 29 factors
• COCOMO 81 - 15 factors
• JPL - 40 factors

4
Cost Adjustment Factors

• TRW
• Easy 0.8
• Medium 1.0
• Hard 1.2
• New module 1.0
• Old module 0.7

5
Development Efforts (MM)

• IBM-FSD W 5.2 L0.91
• RADC W 4.86 L0.976
• Doty W 5.25 L1.057
• JPL W 2.43 L0.962

6
Staffing Size (Persons)

• IBM-FSD S 0.409 W0.65
• RADC S 0.388 W0.641

7
Project Duration (Months)

• IBM-FSD T 2.47 W0.35
• RADC T 3.59 W0.358
• T 4.55 L0.349

8
COnstructive COst Model
Barry Boehm Software Engineering
Economics Prentice Hall, 1981
9
Basic COCOMO 81

• Organic MM 2.4 (KDSI)1.05
• TDEV 2.5 (MM)0.38
• Semidetached MM 3.0 (KDSI)1.12
• TDEV 2.5 (MM)0.35
• Embedded MM 3.6 (KDSI)1.20
• TDEV 2.5 (MM)0.32

10
Intermediate COCOMO 81

• Organic (MM)nom 3.2 (KDSI)1.05
• TDEV 2.5 (MM)0.38
• Semidetached (MM)nom 3.0 (KDSI)1.12
• TDEV 2.5 (MM)0.35
• Embedded (MM)nom 2.8 (KDSI)1.20
• TDEV 2.5 (MM)0.32

11
COCOMO 81
12
COCOMO 81
13
COCOMO 81
14
COCOMO II, 1997 Challenges faced in calibrating
COCOMO II GUI builders, COTS, 4GLs, reuse Need
to rethink size metrics Distributed interactive
applications Web- based, object- oriented,
event- based Middleware effects New process
models (evolutionary, incremental, spiral)
Phases overlap Where are cost measurement
endpoints? Lack of good data not enough data
(i. e. very little degrees of freedom) lack of
dispersion heteroskedasticity
15
COCOMO II The 1997 version Multivariate
Linear Regression with 10weighted average of
expert- determined and data-determined The 1998
version Bayesian Regression Analysis Model
more Data- Determined The 19??/ 20?? version
100 Data- Determined
16
COCOMO II COCOMO II. 1997 Calibration Process
Began with expert- determined a- priori model
parameters Iterated with Affiliates (Result gt
Original Post Architecture Model) - Collected
Data - Identified and consolidated highly
correlated model parameters - Statistically
determined estimates of consolidated model
parameters from data Using logarithms to
linearize regression - Used data determined model
parameters to adjust a- priori model parameters
Experimented with weighting factors
17
COCOMO II
18
COCOMO II
19
COCOMO II Consolidated Highly
Correlated Parameters TIME 1.0000 0.6860 -0.2855
-0.2015 STOR 0.6860 1.0000 -0.0769 -0.0027 ACAP
-0.2855 -0.0769 1.0000 0.7339 PCAP -0.2015
-0.0027 0.7339 1.0000 TIME STOR ACAP PCAP What
do we do? Þ Combine TIME STOR to give RCON
(Resource Constraints) ACAP PCAP to give PERS
(Personnel Factors) Thus, 15 effort multipliers
instead of 17 for calibration
20

• COCOMO II
• Process Maturity does effect effort. A one
  increment change in PMAT (Level 1 Upper to Level
  2, Level 2 to
• Level 3, etc.) results in a 7 to 21 reduction
  in effort for a 30 KDSI project.
• Effect is larger for larger products
• Using it as a scale factor appears to provide a
  stronger influence on effort than as a
  multiplicative factor.
• Its influence is less than the personnel
  capability of the team, about the same as product
  complexity (CPLX), and higher than other COCOMO
  cost drivers.
• Process Maturity should be in all Software Cost
• Estimation Models it is well defined and
  measurable.
• Some Observations on Effects of Process Maturity
  on Effort

21
COCOMO II
22
COCOMO II
23
COCOMO II